Extract
File in. Cited JSON out.
POST a PDF/JPG/PNG with your JSON/XML schema. Grammar-constrained generation plus validators return ≥99% schema-valid output — every value with confidence and bbox.
Three capabilities, one evidence format
Pick a capability to see what it returns in Fieldry.
invoice_no · INV-2026-0410.99
total · 1.240,500.97
vat → nullmissing
Schema-constrained Extract
Grammar-constrained generation plus format validators. ≥99% schema-valid output — no repair loops, no drift.
- Any JSON/XML schema + tables
- Null-honesty, never invented
- Sync ?wait or async 202
Extraction
Dumps. Cited JSON.
What changes when schema, evidence and honesty are built into the API.
After emphasis: 50%
Text without fields or types
Schema breaks on edge cases
Auditor asks: where from?
+50% extra for scores
Schema-valid JSON
Fieldry
Evidence on every value
Fieldry
Null instead of invented
Fieldry
Credits in header
Fieldry
Send a file plus a JSON or XML schema. Get the same schema back filled: value, confidence, status and evidence (page, text, bbox) per field. Missing fields return null — never invented.
PDF/JPG/PNG up to 500 pages. Any custom schema, tables included. Large schemas work; dense pages fall back to Deep Extract behavior without a surcharge tier.
OCR returns raw text. ChatGPT returns plausible text. Fieldry returns your exact schema with validators and per-field evidence — plus X-Credits-Used on every response.
Parse your first document today
Schema in, cited JSON out.
An invoice lands as a phone photo. Fieldry returns the schema with page and bbox evidence — no template, no repair loop.